Gridiron · Game Center · 2022
The Fourth Down Room
Go, kick or punt — and what the win probability said.
Went for it on 29.9% of 137 fourth downs where the rest of the league goes 19.6%.
How the baseline is built · no recommendation engine ▸
There is no recommendation engine here. Scoring the choice a coach did NOT make would mean inventing a win-probability figure for a play that never happened, so the baseline is the league’s own revealed behaviour instead — what all 32 teams actually did from the same spot, 2018–2026.
3,906 real fourth-down decisions in the 2022 regular season. 217 more snaps are set aside — penalties that wiped the play before it resolved, and victory-formation kneels — because neither one decided anything.
Every fourth down. Every regular-season fourth down that resolved into a real decision. 32 of 32 teams clear 20 decisions in every state on the field; a grid cell under 25 decisions is withheld rather than painted.
There is no recommendation engine here. Scoring the choice a coach did NOT make would mean inventing a win-probability figure for a play that never happened, so the baseline is the league’s own revealed behaviour instead — what all 32 teams actually did from the same spot, 2018–2026.
Painted after the last game's play-by-play landed · the room repaints as each game's play-by-play lands, usually within two hours of the final, or the next morning when its release runs late
Reading today’s slate…
The signature object · and the filter
The decision grid
Pooled 2018–2026, so every cell is thousands of decisions rather than a season of noise. Read the middle of the top row before trusting anything else: the league goes for it LESS at the opponent's 16-25 than at the 26-38, because 16-25 is a comfortable field goal and 26-38 is a long one. A grid that fell smoothly with field position would be a grid that had lost the plot.
League 2018–2026
| To gain | |||||||
|---|---|---|---|---|---|---|---|
n/aFourth and 7-10 from the Goal line is not a possible state. | |||||||
n/aFourth and 11+ from the Goal line is not a possible state. |
Numeral = go-rate · small figure = decisions in the cell · n/a = a state the field cannot produce
League 2018–2026 · every state on the field
The call sheet
The bar is what coaches did. The rule across it is what this room's own win-probability means say they should have done: on every cell where all three choices cleared 25 plays, the model prefers going when going returned at least as much win probability as the better of kicking and punting, weighted by the decisions actually made there. Neither number is a forecast of a road not taken — both are measured on the plays themselves.
On fourth and 4-6 the league went for it 14.6% of the time while its own returns favoured going on 100.0% of the same 901 snaps — the widest gap on the sheet.
Thin sample: 5 qualified · too few rows to read as a set
Go over expected · every state on the field
The aggression ladder
Raw go-rate lies about who is aggressive, because it does not know which spots a team stood in. Go over expected does: it subtracts the league's own pooled rate in the SAME cells, weighted by how often this team was there. The two orderings are not the same list, and where a team moves between them is the finding.
Decisions made
- 1
ArizonaCardinals+10.430% of 137Arizona Cardinals, go over expected 10.4 points, 30 percent of 137 decisions - 2
ClevelandBrowns+8.431% of 135Cleveland Browns, go over expected 8.4 points, 31 percent of 135 decisions - 3
PhiladelphiaEagles+6.629% of 111Philadelphia Eagles, go over expected 6.6 points, 29 percent of 111 decisions - 4
Green BayPackers+5.528% of 110Green Bay Packers, go over expected 5.5 points, 28 percent of 110 decisions - 5
ChicagoBears+5.421% of 117Chicago Bears, go over expected 5.4 points, 21 percent of 117 decisions - 6
DetroitLions+4.232% of 117Detroit Lions, go over expected 4.2 points, 32 percent of 117 decisions - 7
New YorkJets+1.819% of 145New York Jets, go over expected 1.8 points, 19 percent of 145 decisions - 8
HoustonTexans+1.718% of 143Houston Texans, go over expected 1.7 points, 18 percent of 143 decisions - 9
MiamiDolphins+1.321% of 117Miami Dolphins, go over expected 1.3 points, 21 percent of 117 decisions - 10
Los AngelesRams+1.116% of 120Los Angeles Rams, go over expected 1.1 points, 16 percent of 120 decisions - 11
SeattleSeahawks+0.717% of 121Seattle Seahawks, go over expected 0.7 points, 17 percent of 121 decisions - 12
Los AngelesChargers−0.321% of 132Los Angeles Chargers, go over expected -0.3 points, 21 percent of 132 decisions - 13
WashingtonCommanders−0.718% of 137Washington Commanders, go over expected -0.7 points, 18 percent of 137 decisions - 14
CarolinaPanthers−1.016% of 132Carolina Panthers, go over expected -1.0 points, 16 percent of 132 decisions - 15
MinnesotaVikings−1.116% of 121Minnesota Vikings, go over expected -1.1 points, 16 percent of 121 decisions - 16
TennesseeTitans−1.213% of 128Tennessee Titans, go over expected -1.2 points, 13 percent of 128 decisions - 17
AtlantaFalcons−1.416% of 110Atlanta Falcons, go over expected -1.4 points, 16 percent of 110 decisions - 18
IndianapolisColts−1.519% of 135Indianapolis Colts, go over expected -1.5 points, 19 percent of 135 decisions - 19
JacksonvilleJaguars−1.523% of 117Jacksonville Jaguars, go over expected -1.5 points, 23 percent of 117 decisions - 20
DenverBroncos−1.616% of 153Denver Broncos, go over expected -1.6 points, 16 percent of 153 decisions - 21
BuffaloBills−1.915% of 86Buffalo Bills, go over expected -1.9 points, 15 percent of 86 decisions - 22
Las VegasRaiders−1.921% of 116Las Vegas Raiders, go over expected -1.9 points, 21 percent of 116 decisions - 23
New YorkGiants−2.017% of 126New York Giants, go over expected -2.0 points, 17 percent of 126 decisions - 24
Tampa BayBuccaneers−2.420% of 142Tampa Bay Buccaneers, go over expected -2.4 points, 20 percent of 142 decisions - 25
DallasCowboys−2.916% of 116Dallas Cowboys, go over expected -2.9 points, 16 percent of 116 decisions - 26
BaltimoreRavens−5.417% of 116Baltimore Ravens, go over expected -5.4 points, 17 percent of 116 decisions - 27
San Francisco49ers−5.516% of 108San Francisco 49ers, go over expected -5.5 points, 16 percent of 108 decisions - 28
PittsburghSteelers−6.212% of 125Pittsburgh Steelers, go over expected -6.2 points, 12 percent of 125 decisions - 29
CincinnatiBengals−6.912% of 98Cincinnati Bengals, go over expected -6.9 points, 12 percent of 98 decisions - 30
New OrleansSaints−8.29% of 118New Orleans Saints, go over expected -8.2 points, 9 percent of 118 decisions - 31
Kansas CityChiefs−9.413% of 91Kansas City Chiefs, go over expected -9.4 points, 13 percent of 91 decisions - 32
New EnglandPatriots−9.410% of 126New England Patriots, go over expected -9.4 points, 10 percent of 126 decisions
Bar diverges from the league rate · under 20 decisions the figure is withheld, not guessed
2022 · every state on the field
What each choice actually returned
Mean win probability added, measured on the plays themselves — not a forecast of the road not taken. Every figure below is withheld under 25 plays, because the rare choice in a cell is rare precisely because it is unusual, and its mean is the loudest number in the row for exactly that reason.
Went for it
73519% of choices
Mean WP added+0.001
Measured on 480 of them — the rest sit in cells too thin to average.
Converted56.9%
Drive ended in points41.7%
Kicked
97125% of choices
Mean WP added+0.003
Measured on 887 of them — the rest sit in cells too thin to average.
Made85.9%
Punted
2,20056% of choices
Mean WP added+0.007
Measured on 2,177 of them — the rest sit in cells too thin to average.